JP6313929B2 - Method and system for monitoring structures - Google Patents

Description

The present invention relates to a method and system for monitoring structures, and more particularly to predicting fatigue and damage resistance. The present invention is applicable to any plant structure, wind turbine, ship, building, bridge, tower and preferably (optionally ) an aircraft.

  Structural fatigue can be defined as material damage due to the gradual growth of microcracks under conditions of cyclic loading. Generally speaking, the fatigue life of a plant structure is the time taken for damage in a specific environment where a repeated load is applied. Fatigue life consumption assessment is not only an important part of plant design and measurement, but also the operational life of the plant must be managed by a so-called structural fatigue monitoring system.

  A number of monitoring systems have been considered for assessing the fatigue life consumption of aircraft structures. Such systems have traditionally been used for military aircraft, but have recently been used for specific applications in commercial aircraft. These systems have two major advantages. That is, ensuring the safe operation of the aircraft and reducing the cost of ownership by optimizing the use and maintenance work of the aircraft over its entire operating life.

  Most fatigue monitoring systems have several characteristics, which are a group of fatigue monitoring systems according to three main characteristics: system philosophy, method criteria and application concepts. Allows classification.

  The basic idea of the system is to define the scope (functional range) of the system. The fatigue monitoring system is divided into two groups of damage detection and damage prediction according to the basic idea. The purpose of the damage detection group's system is to identify and measure the location and severity of accidental damage (due to structural fatigue, or other causes such as corrosion and accidents). On the other hand, systems belonging to the damage prediction category are selected from a predetermined set and estimate the location and / or severity of possible damage taking into account the particular aircraft application.

  Method criteria determine what variables the system should use for damage detection or prediction. The two main groups can be defined as direct methods and parametric methods. A direct approach system (direct system) measures directly in a structure some physical variable that can be used without the aid of an external model. This is an induction because the system derives generic assumptions from a particular set of data. For example, the system has a number of strain sensors that measure strain at several locations in the structure, and that information can be used to calculate fatigue and damage resistance. A parametric approach system (parametric system) uses the general operating parameters of an aircraft to input into a specific model to obtain the necessary data. This is a deductive method because it builds certain assumptions about the structure based on general measurements. For example, the flight cycle and time of flight can be used to manage aircraft usage and to apply a maintenance program according to a set of codes relating to the aircraft sortie profile.

  Each method has advantages and disadvantages. The direct system is accurate and precise. This is because the parameters used for position calculation (damage detection system) or crack initiation and crack growth calculation (damage prediction system) are measured directly from the structure. However, the cost required for sensor mounting and maintenance is generally high. Parametric systems, on the other hand, generally lack accuracy and precision due to the need to use external models to obtain useful data (in practice, the biggest problem towards the realization of a reliable parametric system is The challenge is the complexity of building an accurate model for information processing). However, in many cases, parametric systems are less expensive than direct systems because data is acquired from other systems already installed on the aircraft.

  Historically, parametric systems were developed earlier than direct systems because of their simplicity (eg, vertical load ratio excess counter). As technology develops, more and more sophisticated recorders (eg, distortion data recorders) are installed onboard, and direct systems have begun to be used and have been the preferred concept for many years. In the past 20 years, parametric systems have been used again due to improvements in models and processing capabilities for handling parametric data.

There are three application concepts depending on the number of aircraft to be monitored and the time period being tracked.
-Individual Aircraft Tracking (IAT) where each aircraft in the fleet is monitored over the entire operating life
-Temporary Aircraft Tracking (TAT) where a limited number of aircraft are monitored over a limited period of time
-Selective Aircraft Tracking (SAT) where several aircraft in the fleet are monitored over the entire operating life

  The present invention provides a parametric prediction method and parametric prediction system capable of individual aircraft tracking that combines the accuracy and precision of a direct prediction system and the low cost of a conventional parametric prediction system. The present invention applies not only to aircraft, but to any plant structure (eg, wind turbine, ship, building, bridge or tower) where general parametric data may be related to the strain level at one or more locations in the structure. Is also applicable.

  The above problems can be solved by the method of claim 1 and the system of claim 14. The dependent claims define preferred examples of the invention.

In a first aspect, a method for monitoring a structure according to the present invention comprises:
A plurality of operating parameters x _i (t _j ) and at least one distortion data y (t _j ) acquired synchronously in time series (t ₁ , t ₂ ,... T _m ) using the same or different sampling rates. And an acquisition step of acquiring data having
A construction step of constructing a feature point data set from the data obtained in the obtaining step;
In order to perform the supervised function approximation method of non-adaptive prediction, the operation parameter x _i (t _j ) and the distortion data y (t _j ) are calculated using the constructed data set of the feature points. A modeling step for modeling the relationship,
The construction step of constructing the data set of the feature points includes:
For each time sample t _j , j = 1, ..., m, the data specified as follows
In and, x _i ^min, x _i ^max, respectively y ^min and y ^max, the time series _{(t 1, t 2 ...,} t m) the minimum value of x _i in the maximum value of x _i, the minimum value of y And the maximum value of y. )
The data vector
Sorting the data into N _R groups containing data each group having an absolute value within a predetermined range according to the absolute value of the data vector;
An erasing step of calculating an angle α between the first vector and the second vector in each group, and erasing either the first vector or the second vector when the angle α is equal to or smaller than a predetermined angle θ. Is included.

The strain data y (t _j ) can be acquired at one or more structural positions of the structure.

  In a preferred example of the above method, the learning of the non-adaptive prediction supervised function approximation method is performed by an ANN (artificial neural network). However, other methods such as state space models, polynomials, or autoregressive models can be used.

In the example method, the data vector having minimum and maximum distortion values in the respective groups in the erasing step.
Remains.

  In the above method example, when each group is associated with the minimum number of points of the group, and the number of the feature points remaining in one group after the erasing step is less than the minimum number of points, The one group is divided into two groups of the same size, and the erasing step is repeated until the minimum number of feature points remains in the original one group.

In the above method example, prior to the construction step of constructing the data set of the feature points, a pre-processing is performed on the acquired data (x _i , y), and the pre-processing step of performing the pre-processing includes: , Preferably including at least one selected from applications of error detection and error correction algorithms, normalization, filtering, resampling, and generation of low frequency data sets. In a preferred example, the preprocessing step of performing the preprocessing on the acquired data (x _i , y) includes generating the low-frequency data set, and the data set of the feature points includes Constructed from a low frequency data set.

The method, the use and the being the acquired not listed in data set data of feature points, the feature points wherein the selected from the data set is arbitrary characteristic points, the operating parameters x _i A verification step of verifying the modeled relationship between (t _j ) and the distortion data y (t _j ) may be included. In this example, the method is preferably obtained by applying an unreduced original low-frequency data set to the modeled relationship after learning and verification, and the applying step. and including an adjustment coefficient calculation step of calculating an adjustment coefficient as a ratio of the model output.

  In one example, the method includes the acquired driving that has undergone the learning and the verification, the acquired operating parameter, the calculated adjustment factor, and optionally the pre-processing. And a history calculation step of calculating a history of distortion of at least one structure based on the parameters. The method may include an additional preprocessing step for preprocessing the acquired parameters.

  In one example, the method includes a correction distortion calculating step of calculating correction distortion using the adjustment coefficient.

In one example, the acquired operating parameters are pre-processed, and the pre-processing step includes a generation step of generating distortion of the original full band, and the method further includes: A test step of checking the operation of the continuous system in operation by comparing the strain with the calculated strain. The method can include a corrective task execution step of executing a corrective task if the inspected continuous system operation in operation indicates poor results. Correction tasks may include updating modeled relationships and / or updating feature point datasets.

In one example, the method includes a consumption calculation step of calculating fatigue life and damage resistance consumption of at least one structure using the calculated strain.

  The acquired data and / or the calculated output can be stored in a database.

In a preferred example, the method is applied to an aircraft structure, the operating parameters x _i (t _j ) are obtained from N aircraft, and the distortion data y (t _j ) is obtained from M reference aircraft. Acquired, M is N or less. Preferably, M is about 10% to 20% of N. The strain data y (t _j ) can be acquired at one structural location of each reference aircraft, and preferably is acquired at more than one location of each reference aircraft.

  In a second aspect of the invention, the present invention is a system for monitoring a structure, wherein at least one physical strain sensor is mounted in a selected position of the structure in at least one structure. Defines a system having an acquisition and recording device adapted to collect operating parameters from a processor and a processing unit adapted to perform the steps according to the first aspect of the invention.

  In a preferred example, the structure to be monitored is at least one aircraft. In this case, the acquisition / recording device preferably collects the operating parameters of the aircraft from the aircraft and, if the at least one physical strain sensor is mounted on a reference aircraft, the strain from the physical sensor. A device connected to an aircraft computer and / or a dedicated sensor that collects signals. The processing unit can be implemented in a ground processing device.

  The method and system of the present invention is a virtual strain sensor constructed using a non-adaptive prediction supervised function approximation method (e.g., ANN) for crack initiation and crack growth in a predetermined set of structures. It can be thought of as a set of virtual strain sensors capable of generating a time series of loads that can be used for estimation. In particular, the present invention relates to a method and system that enables the realization of a set of virtual strain sensors that calculate the strain at a selected position of the plant structure from normal operating data of the plant structure. The method and system of the present invention completes a maintenance program that takes into account the actual operation of the structure by allowing estimation of the rate of crack growth at and after the occurrence of a crack at a given location of the structure. Provide additional means.

  All features described in this specification (claims, specification, drawings, etc.) and / or all steps of the described method are optional except for mutually exclusive features and / or combinations of steps. Can be combined.

6 is a flowchart illustrating data redundancy removal steps according to a preferred embodiment of the method of the present invention. It is a figure which shows the state by which the sorting process of a highly redundant data set and the sorted data were divided | segmented into the group which has a similar absolute value. It is a figure which shows an example of the autonomous monovalent | monohydric function (it is time-independent) for clarifying the relationship between the parameter and structure distortion used as modeling object. 3 is a flowchart according to a preferred embodiment of the method of the present invention. 6 is a flowchart including steps of fatigue life prediction and crack growth calculation using a load time series obtained from a virtual strain sensor based on an ANN (artificial neural network), according to an embodiment of the method of the present invention. It is a figure which shows the example of the data reduction process for the removal of redundant data. It is a figure which shows another example of the data reduction process for the removal of redundant data. Fig. 6 is a flow chart including the steps of continuous system self-verification according to a preferred embodiment of the method.

  For a better understanding of the present invention and its objects and advantages, the drawings are appended hereto.

1, 4 , 5 and 7 are flowcharts illustrating a method for monitoring a structure according to a preferred embodiment of the present invention.

  Although the method is described as being applied to aircraft monitoring, it is also applicable to other structures such as bridges or ships, for example. Depending on the structure being monitored, different operating parameters (i.e. parameters known to affect the distortion of the structure) will be used.

The method starts with a group of parameter signal data x _i recorded synchronously and data y containing distortion signal data in one flight or multiple flights (see FIG. 1). . This data set is acquired and downloaded to the processing device. In the processing apparatus, preprocessing including error detection and error correction, filtering, and resampling steps is performed on the data set. As a result, a so-called low frequency data set consisting of correction signals each having the same sampling rate and band is obtained (see step 22 in FIG. 4 ).

A data set acquired at time (t ₁ , t ₂ ,... T _m ) can be understood as the following multidimensional data set.

Here, the distortion signal y and the parameter signal x _i are connected by a functional relationship, the parameter signal x _i is an independent variable, and the distortion signal y is a dependent variable having the following functional relationship.

The purpose of the method is to remove the redundancy of the data set (if such redundancy exists) to use the smallest amount of data that fits the model. Therefore, the data is a vector to be compared.
It is thought that this is a record.

The preprocessed data set including parameters x _i and distortion signal y is input to the selector (step 10). For each calculation point t _j , the selector associates a parameter signal x _i having a functional relationship with a distortion signal y corresponding to the parameter signal x _i in a functional relationship. At each calculation point t _j , the data is vectorized (step 11), and the vectorized data
Are normalized and their absolute values are evaluated (step 12).

Preferably, in the normalization of the preprocessing step, the maximum and minimum values of x _i and y are specified in order to convert the physical parameter range into the normalized range [−1, 1].
(1) First region with low absolute value and large variation (2) Second region with medium absolute value and small variation (3) Third region with high absolute value and large variation

In order to reduce the number of data by ignoring redundant data, redundancy is evaluated using two predetermined parameters (step 14). The two parameters are the relative multidimensional distance (θ) and the number of calculation groups (N _R ) between two data vectors, the former being used for angle comparison and the latter for absolute value comparison.

Then, as the first filtering for comparing the data vectors, the data is divided into N _R groups each having a similar absolute value, and all the data vectors included in each group are within a specified range. Will have an absolute value (see FIG. 2 ). The number of vectors included in each group is determined by the statistical distribution area. For example, in the first region and the third region in FIG. 2, there are almost no vectors in each group due to a large variation in absolute value. On the other hand, in the second region, since the variation of the absolute value in the region is small, there are many vectors.

When the above operation is performed on all pairs of vectors in each group, only the feature points of each group remain and all redundant data should be erased. The result of the redundancy evaluation is a set of feature point positions in the M data sets based on predetermined parameters (θ, N _R ) (step 15). Then, using this information, a recorded and preprocessed data set, that is, M is acquired as a learning data set composed of feature points (step 16) and a verification data set including redundant information (step 17). Divided for.

FIG. 6 is a diagram illustrating two examples of the configuration of the feature point data set. In FIG. 6 A, the density of points is shown a two-dimensional data set including a high region. In this example, the variables x ₁ and x ₂ are two flight parameters (Mach number and vertical load ratio, respectively). With respect to a specific accurate circle Δ, the center of gravity (G) of the area where the density of the points is high is determined, and all the points Z in the circle are centered on the center of gravity (G) with a predetermined accuracy indicated by the circle shown It is thought that it is represented by the feature point located. This accuracy, when fitted the feature point G in one model are those suitable in terms of points in the circle can be interpolated with predetermined accuracy, thereby, a characteristic point X ₁ feature point G Represents the portion of the circle that contains it, and other points within the circle are erased. Points that are outside the portion of the circle (eg, X ₂ ) are maintained. Thus, the feature point X ₁ is a perspective of the feature point is represented based on a (plurality of) point distribution, with a statistical significance.

Figure 6 B is a diagram showing in Kazehai view multidimensional interpretation of the foregoing described example, each direction, variable (e.g., Mach number, altitude, vertical load ratio, engine thrust, the displacement amount of the aileron, elevator Displacement amount, total weight and center of gravity). In this multidimensional representation, each point X is represented by a polygon, and the crown-shaped polygons Δ ₁ -Δ ₂ surrounding the feature point G correspond to the circle in the above-described two-dimensional representation. Therefore, all the polygons X included in the crown are represented by the feature points G, and the feature points G have a statistical meaning of multidimensional distribution.

The parameter x ₁ data (parametric data x ₁ ) can be acquired by a recorder installed in the aircraft in order to acquire appropriate aircraft parameters from other aircraft computers and / or dedicated sensors. Sensors and recorders can be connected using standard data buses and digital transmission protocols (eg, ARINC429, MIL1553B, RS232, etc.), and the recorder preferably has an acquisition interface that can accommodate all such data formats. Have Parametric data needs to be recorded synchronously. Parametric data can be extracted from, for example, existing aircraft computers, engine controls, landing gear brakes and steering devices, aircraft controls, atmospheric data, inertial data, weight and center of gravity or central management. Examples of parametric data include speed, altitude, vertical load ratio, Mach number, and the like.

  Also, the strain data y is acquired by at least one physical strain sensor mounted on a subset of the fleet (preferably a representative aircraft for each of the various structural forms of the fleet). The strain sensor is connected to the recorder in an analog fashion, and the recorder needs to have all the necessary acquisition components for the particular strain sensor used. Alternatively, the strain sensor is digitally connected to the recorder and acquires data from other devices. In this case, the recorder must be compatible with the data transmission protocol of the intermediary device and maintain synchronization with the aircraft parameters.

The main purpose of the above method is a function of the strain y (dependent variable) measured at one or more positions of the structure and the motion parameters x ₁ (eg independent variables such as speed, altitude and vertical load ratio). Is to build a relationship. This functional relationship can be modeled using a non-adaptive supervised function approximation method (eg, ANN) when several conditions are met. It should be understood that while referring to ANN, any other non-adaptive prediction supervised function approximation method may be used.

First, functional relationships must only be evaluated. That is, for each combination of operation parameters, one functional relationship needs to exist, and there is only one distortion value. This is explained with examples in Fig. Suppose that it is desirable to approximate the distortion at several locations of the structure in flight using only one operating parameter x ₁ (eg Mach number (see FIG. 3 (a) )). In this example,

Since up to three distortion values can be seen in each of the above, the functional relationship is

Not only evaluated between. Therefore, in this case, there is no unique functional relationship in the interval. However, if a second parameter x ₂ (for example, altitude (see FIG. 3B )) that is not evaluated in plural in the operating range for the distortion is added to the functional relationship, the altitude and the general purpose of Mach and distortion This relationship is uniquely evaluated ( see FIG. 3 (c) ). Therefore, by introducing sufficient variables in the function relationship, the function relationship can be approximated by an artificial neural network within the operating range of the parameters ( see points A to D in FIG. 3C ).

  Also, the functional relationship must be autonomous and time independent. All strain values must be determined by operating parameters and not by time samples.

  And the sampling rate and bandwidth of the data and parameters and distortion used for fitting to the model must be equal.

  In order to satisfy the above three conditions, each strain measurement point of the structure needs to have a set (group) of basic parameters that give the greatest influence on the functional relationship. The effect of parameters can be grouped by structure area. For example, one of the most important parameters at the base of an aircraft wing is the vertical load ratio measured at the center of gravity, and some of the main load of the fuselage is caused by cabin differential pressure, so these parameters are This is a standard for identifying distortion in a part.

  Therefore, the relationship between parametric data and distortion is modeled using an artificial neural network and learned using a feature point data set.

FIG. 4 shows the process used to fit the model. Starting point of the process is a data acquisition by the recorder mounted on board the aircraft strain gauge is mounted, the data set containing operating parameters x _i and strain gauge signal y is generated. Then, the operational parameters x _i associated with each strain measurement point (along with the strain y itself) are extracted from the data set (step 20) and a subset {x _i, y} is generated. The subset {x _i, y is preprocessed (step 21). The pre-processing preferably includes, for example, error detection and error correction, signal synchronization (ie, delay removal), band matching, collation into high frequency data sets (step 27) and low frequency This includes signal filtering for dividing into data sets (step 22) and signal resampling for matching sampling rates. The low frequency data set includes parameters x _i and distortion y with equal bandwidth and sampling rate, respectively (step 22), and is used for fitting to the artificial neural network (step 25). On the other hand, the high frequency data set includes distortion data of the entire band (step 27) and is used to calculate an adjustment coefficient (step 28).

  Then, the low frequency data set is introduced into the flow chart of the data reduction algorithm (step 22), and the characteristic points of the low frequency data set to be applied to the learning data set, that is, the artificial neural network (step 23), and The redundant information of the verification data set, that is, the low frequency data set is acquired (step 24). In general, the ratio of feature points to non-feature points can be 100 to 1, which facilitates learning of artificial neural networks. The learning data set (from step 23) is used for learning the artificial neural network model (step 25). The model can be validated (step 26) by providing the trained artificial neural network (step 25) with a redundant data set (step 24). If the verification error is similar to the learning error, processing continues and an adjustment factor is calculated (step 28). If the validation error is not similar to the learning error, the learning data and / or model must be changed. An adjustment factor (AF) is obtained (step 28) using the artificial neural network model after learning and verification and the entire band distortion data set (from step 27). The method for obtaining the adjustment factor (AF) is the ratio between the fatigue life calculated from the strain time series of the entire band and the fatigue life calculated from the simulation strain time series obtained from the artificial neural network model. Including the evaluation of Therefore, the adjustment factor (AF) indicates the ratio of fatigue damage between the output of the artificial neural network model (low frequency calculated strain) and the full-band strain. Also, the adjustment factor (AF) can be used for virtual sensor output fatigue calculation to compensate for high frequency and model performance effects. Then, the adjustment coefficient (AF) is incorporated into the artificial neural model (step 29), and the artificial neural model is completed. In this way, the final set of artificial neural models and adjustment factors are ready for implementation in subsequent damage calculation steps.

FIG. 5 is a diagram illustrating an operation of a method including a step of calculating fatigue damage at a specific position according to a preferred embodiment of the present invention. Data including the parametric data x _i is downloaded from the recorder (step 20), and pre-processing (eg, error detection and error correction, synchronization, filtering, and resampling) is performed (step 21). An artificial neural network model after learning and verification (in step 29) with an appropriate adjustment factor (used later) is input with preprocessed data to obtain a time history of calculated distortion values. (Step 30). Further, using the calculated strain (pressure) time history calculated in step 31, fatigue and crack growth at or around the selected position are analyzed (step 32), and an actual adjustment coefficient using the position adjustment factor is analyzed. Fatigue life consumption and / or crack growth is evaluated. This process may be performed for all selected distortion locations of each aircraft in the air fleet.

FIG. 7 illustrates the continuous operation of a model based on an artificial neural model (so-called continuous self It is a figure which shows the method including the step of verification.

The flowchart (gray part) in the left part is substantially the same as FIG. 5 , that is, the workflow during operation of the system. The right part of the flowchart (black part) contains the system self-verification. Acquired data including parametric data x _i and distortion data y is downloaded (step 20), and distortion sensor signal y is extracted. The distortion sensor signal y is subjected to preprocessing (for example, error detection, error correction, and filtering) (step 21). All-band distortion data is output by the preprocessing (step 27), and an actual distortion time history is acquired using the all-band distortion data (step 40). Then, an actual distortion time series is obtained from the distortion measurement (step 40), and this actual distortion time series is compared with the calculated distortion time series (in step 31) calculated by the artificial neural network, and the adjustment coefficient is used. (Comparison can be made with respect to time series, strain spectra obtained by rainflow analysis, fatigue life, etc.). By checking the comparison result, it is determined whether or not it is within a predetermined margin (step 41). If the result of this comparison (in step 41) is within a predetermined margin, it is determined that the model based on the artificial neural network is valid (step 43). If the result of the comparison (in step 41) is not within a predetermined margin, the model based on the artificial neural network needs to be updated according to the model calculation procedure (step 42).

Claims (15)

A method for monitoring a structure,
A plurality of operation parameters x _i (t _j ) and at least one distortion data y (t _j ) acquired synchronously at time series t _j , j = 1,..., _M , using the same or different sampling rates. And an acquisition step of acquiring data having
A construction step of constructing a feature point data set from the data obtained in the obtaining step;
In order to perform the supervised function approximation method of non-adaptive prediction, the operation parameter x _i (t _j ) and the distortion data y (t _j ) are calculated using the constructed data set of the feature points. A modeling step for modeling the relationship,
The construction step of constructing the data set of the feature points includes:
For each time sample t _j , j = 1, ..., m, the data specified as follows
In and, x _i ^min, x _i ^max, respectively y ^min and y ^max, the time series _{(t 1, t 2 ...,} t m) the minimum value of x _i in the maximum value of x _i, the minimum value of y And the maximum value of y. )
The data vector
Sorting the data into N _R groups containing data each group having an absolute value within a predetermined range according to the absolute value of the data vector;
An erasing step of calculating an angle α between the first vector and the second vector in each group, and erasing either the first vector or the second vector when the angle α is equal to or smaller than a predetermined angle θ. And a method comprising. 2. The method of claim 1, wherein, in the erasing step, the data vector having minimum and maximum distortion values in the respective groups.
How to remain.   3. The method according to claim 1, wherein a minimum number of points of the group is associated with each group, and the number of feature points remaining in one group after the erasing step is less than the minimum number of points. In this case, the one group is divided into two groups having the same size, and the erasing step is repeated until the minimum number of feature points remains in the original one group. 4. The method according to any one of claims 1 to 3, wherein said method further comprises a step preceding said acquired data (x _i , y) prior to said construction step of constructing said data set of said feature points. A preprocessing step for performing processing, preferably at least one selected from applications of error detection and error correction algorithms, normalization, filtering, resampling, and generation of low frequency data sets. A method comprising a preprocessing step comprising. 5. The method of claim 4, wherein the preprocessing step of performing the preprocessing on the acquired data (x _i , y) includes generating the low-frequency data set, and the data set of the feature points Is constructed from the low-frequency data set. 6. A method as claimed in any preceding claim, further comprising selecting from the acquired data not included in the data set of the feature points and optionally but from the data set of the feature points. A verification step of verifying the modeled relationship between the operating parameter x _i (t _j ) and the distortion data y (t _j ) using the characterized feature points.   7. The method of claim 6, when dependent on claim 5, further comprising: providing the original modeled low frequency data set to the modeled relationship after learning and the validation. And an adjustment coefficient calculation step of calculating an adjustment coefficient as a ratio of the model output obtained by the assigning step.   The method according to claim 7, wherein the method further includes the model after the learning and the verification, the acquired operating parameter, the calculated adjustment factor, and optionally the preprocessing. A history calculation step of calculating a history of distortion of at least one structure based on the obtained operating parameters.   9. The method according to claim 8, wherein the method includes a correction distortion calculation step of calculating correction distortion using the adjustment coefficient. The method according to claim 9, wherein the obtained pre-processing parameter is subjected to the pre-processing, and the pre-processing step includes a generation step of generating distortion of the original entire band, and the method further includes: A method comprising an inspection step of examining the operation of a continuous system in operation by comparing the distortion of the entire band with the calculated distortion.   10. The method of claim 9, further comprising a wear calculation step of calculating fatigue life and damage resistance wear of at least one structure using the calculated strain.   12. The method according to claim 1, wherein learning of the non-adaptive prediction supervised function approximation method is performed by an artificial neural network. 13. A method according to any one of the preceding claims applied to an aircraft structure, wherein the operating parameters x _i (t _j ) are obtained from N aircraft and the distortion data y (t _j ) is M pieces. And M is less than or equal to N, preferably about 10% to 20% of N. A system for monitoring structures,
An acquisition and recording device adapted to collect operational parameters in at least one structure from at least one physical strain sensor mounted at a selected location of the structure;
14. A system comprising a processing unit adapted to perform the steps according to the method of any of claims 1-13.   15. The system of claim 14, wherein the structure is an aircraft, the acquisition and recording device is adapted to collect operational parameters of the aircraft from the aircraft, and the at least one physical strain sensor is the A system installed at a selected location on an aircraft structure.